2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/gfp.h>
20 #include <linux/slab.h>
21 #include <linux/blkdev.h>
22 #include <linux/writeback.h>
23 #include <linux/pagevec.h>
25 #include "transaction.h"
26 #include "btrfs_inode.h"
27 #include "extent_io.h"
30 static u64 entry_end(struct btrfs_ordered_extent *entry)
32 if (entry->file_offset + entry->len < entry->file_offset)
34 return entry->file_offset + entry->len;
37 static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
40 struct rb_node ** p = &root->rb_node;
41 struct rb_node * parent = NULL;
42 struct btrfs_ordered_extent *entry;
46 entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
48 if (file_offset < entry->file_offset)
50 else if (file_offset >= entry_end(entry))
56 rb_link_node(node, parent, p);
57 rb_insert_color(node, root);
61 static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
62 struct rb_node **prev_ret)
64 struct rb_node * n = root->rb_node;
65 struct rb_node *prev = NULL;
67 struct btrfs_ordered_extent *entry;
68 struct btrfs_ordered_extent *prev_entry = NULL;
71 entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
75 if (file_offset < entry->file_offset)
77 else if (file_offset >= entry_end(entry))
85 while(prev && file_offset >= entry_end(prev_entry)) {
89 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
91 if (file_offset < entry_end(prev_entry))
97 prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
99 while(prev && file_offset < entry_end(prev_entry)) {
100 test = rb_prev(prev);
103 prev_entry = rb_entry(test, struct btrfs_ordered_extent,
111 static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
113 if (file_offset < entry->file_offset ||
114 entry->file_offset + entry->len <= file_offset)
119 static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
122 struct rb_root *root = &tree->tree;
123 struct rb_node *prev;
125 struct btrfs_ordered_extent *entry;
128 entry = rb_entry(tree->last, struct btrfs_ordered_extent,
130 if (offset_in_entry(entry, file_offset))
133 ret = __tree_search(root, file_offset, &prev);
141 /* allocate and add a new ordered_extent into the per-inode tree.
142 * file_offset is the logical offset in the file
144 * start is the disk block number of an extent already reserved in the
145 * extent allocation tree
147 * len is the length of the extent
149 * This also sets the EXTENT_ORDERED bit on the range in the inode.
151 * The tree is given a single reference on the ordered extent that was
154 int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
157 struct btrfs_ordered_inode_tree *tree;
158 struct rb_node *node;
159 struct btrfs_ordered_extent *entry;
161 tree = &BTRFS_I(inode)->ordered_tree;
162 entry = kzalloc(sizeof(*entry), GFP_NOFS);
166 mutex_lock(&tree->mutex);
167 entry->file_offset = file_offset;
168 entry->start = start;
170 entry->inode = inode;
172 /* one ref for the tree */
173 atomic_set(&entry->refs, 1);
174 init_waitqueue_head(&entry->wait);
175 INIT_LIST_HEAD(&entry->list);
176 INIT_LIST_HEAD(&entry->root_extent_list);
178 node = tree_insert(&tree->tree, file_offset,
181 printk("warning dup entry from add_ordered_extent\n");
184 set_extent_ordered(&BTRFS_I(inode)->io_tree, file_offset,
185 entry_end(entry) - 1, GFP_NOFS);
187 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
188 list_add_tail(&entry->root_extent_list,
189 &BTRFS_I(inode)->root->fs_info->ordered_extents);
190 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
192 mutex_unlock(&tree->mutex);
198 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
199 * when an ordered extent is finished. If the list covers more than one
200 * ordered extent, it is split across multiples.
202 int btrfs_add_ordered_sum(struct inode *inode,
203 struct btrfs_ordered_extent *entry,
204 struct btrfs_ordered_sum *sum)
206 struct btrfs_ordered_inode_tree *tree;
208 tree = &BTRFS_I(inode)->ordered_tree;
209 mutex_lock(&tree->mutex);
210 list_add_tail(&sum->list, &entry->list);
211 mutex_unlock(&tree->mutex);
216 * this is used to account for finished IO across a given range
217 * of the file. The IO should not span ordered extents. If
218 * a given ordered_extent is completely done, 1 is returned, otherwise
221 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
222 * to make sure this function only returns 1 once for a given ordered extent.
224 int btrfs_dec_test_ordered_pending(struct inode *inode,
225 u64 file_offset, u64 io_size)
227 struct btrfs_ordered_inode_tree *tree;
228 struct rb_node *node;
229 struct btrfs_ordered_extent *entry;
230 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
233 tree = &BTRFS_I(inode)->ordered_tree;
234 mutex_lock(&tree->mutex);
235 clear_extent_ordered(io_tree, file_offset, file_offset + io_size - 1,
237 node = tree_search(tree, file_offset);
243 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
244 if (!offset_in_entry(entry, file_offset)) {
249 ret = test_range_bit(io_tree, entry->file_offset,
250 entry->file_offset + entry->len - 1,
253 ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
255 mutex_unlock(&tree->mutex);
260 * used to drop a reference on an ordered extent. This will free
261 * the extent if the last reference is dropped
263 int btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
265 struct list_head *cur;
266 struct btrfs_ordered_sum *sum;
268 if (atomic_dec_and_test(&entry->refs)) {
269 while(!list_empty(&entry->list)) {
270 cur = entry->list.next;
271 sum = list_entry(cur, struct btrfs_ordered_sum, list);
272 list_del(&sum->list);
281 * remove an ordered extent from the tree. No references are dropped
282 * but, anyone waiting on this extent is woken up.
284 int btrfs_remove_ordered_extent(struct inode *inode,
285 struct btrfs_ordered_extent *entry)
287 struct btrfs_ordered_inode_tree *tree;
288 struct rb_node *node;
290 tree = &BTRFS_I(inode)->ordered_tree;
291 mutex_lock(&tree->mutex);
292 node = &entry->rb_node;
293 rb_erase(node, &tree->tree);
295 set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
297 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
298 list_del_init(&entry->root_extent_list);
299 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
301 mutex_unlock(&tree->mutex);
302 wake_up(&entry->wait);
306 int btrfs_wait_ordered_extents(struct btrfs_root *root)
308 struct list_head splice;
309 struct list_head *cur;
310 struct btrfs_ordered_extent *ordered;
313 INIT_LIST_HEAD(&splice);
315 spin_lock(&root->fs_info->ordered_extent_lock);
316 list_splice_init(&root->fs_info->ordered_extents, &splice);
317 while(!list_empty(&splice)) {
319 ordered = list_entry(cur, struct btrfs_ordered_extent,
321 list_del_init(&ordered->root_extent_list);
322 atomic_inc(&ordered->refs);
323 inode = ordered->inode;
326 * the inode can't go away until all the pages are gone
327 * and the pages won't go away while there is still
328 * an ordered extent and the ordered extent won't go
329 * away until it is off this list. So, we can safely
330 * increment i_count here and call iput later
332 atomic_inc(&inode->i_count);
333 spin_unlock(&root->fs_info->ordered_extent_lock);
335 btrfs_start_ordered_extent(inode, ordered, 1);
336 btrfs_put_ordered_extent(ordered);
339 spin_lock(&root->fs_info->ordered_extent_lock);
341 spin_unlock(&root->fs_info->ordered_extent_lock);
346 * Used to start IO or wait for a given ordered extent to finish.
348 * If wait is one, this effectively waits on page writeback for all the pages
349 * in the extent, and it waits on the io completion code to insert
350 * metadata into the btree corresponding to the extent
352 void btrfs_start_ordered_extent(struct inode *inode,
353 struct btrfs_ordered_extent *entry,
356 u64 start = entry->file_offset;
357 u64 end = start + entry->len - 1;
360 * pages in the range can be dirty, clean or writeback. We
361 * start IO on any dirty ones so the wait doesn't stall waiting
362 * for pdflush to find them
364 btrfs_fdatawrite_range(inode->i_mapping, start, end, WB_SYNC_NONE);
366 wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
371 * Used to wait on ordered extents across a large range of bytes.
373 void btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
378 struct btrfs_ordered_extent *ordered;
380 if (start + len < start) {
381 orig_end = INT_LIMIT(loff_t);
383 orig_end = start + len - 1;
384 if (orig_end > INT_LIMIT(loff_t))
385 orig_end = INT_LIMIT(loff_t);
389 /* start IO across the range first to instantiate any delalloc
392 btrfs_fdatawrite_range(inode->i_mapping, start, orig_end, WB_SYNC_NONE);
394 btrfs_wait_on_page_writeback_range(inode->i_mapping,
395 start >> PAGE_CACHE_SHIFT,
396 orig_end >> PAGE_CACHE_SHIFT);
400 ordered = btrfs_lookup_first_ordered_extent(inode, end);
404 if (ordered->file_offset > orig_end) {
405 btrfs_put_ordered_extent(ordered);
408 if (ordered->file_offset + ordered->len < start) {
409 btrfs_put_ordered_extent(ordered);
412 btrfs_start_ordered_extent(inode, ordered, 1);
413 end = ordered->file_offset;
414 btrfs_put_ordered_extent(ordered);
415 if (end == 0 || end == start)
419 if (test_range_bit(&BTRFS_I(inode)->io_tree, start, orig_end,
420 EXTENT_ORDERED | EXTENT_DELALLOC, 0)) {
421 printk("inode %lu still ordered or delalloc after wait "
422 "%llu %llu\n", inode->i_ino,
423 (unsigned long long)start,
424 (unsigned long long)orig_end);
430 * find an ordered extent corresponding to file_offset. return NULL if
431 * nothing is found, otherwise take a reference on the extent and return it
433 struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
436 struct btrfs_ordered_inode_tree *tree;
437 struct rb_node *node;
438 struct btrfs_ordered_extent *entry = NULL;
440 tree = &BTRFS_I(inode)->ordered_tree;
441 mutex_lock(&tree->mutex);
442 node = tree_search(tree, file_offset);
446 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
447 if (!offset_in_entry(entry, file_offset))
450 atomic_inc(&entry->refs);
452 mutex_unlock(&tree->mutex);
457 * lookup and return any extent before 'file_offset'. NULL is returned
460 struct btrfs_ordered_extent *
461 btrfs_lookup_first_ordered_extent(struct inode * inode, u64 file_offset)
463 struct btrfs_ordered_inode_tree *tree;
464 struct rb_node *node;
465 struct btrfs_ordered_extent *entry = NULL;
467 tree = &BTRFS_I(inode)->ordered_tree;
468 mutex_lock(&tree->mutex);
469 node = tree_search(tree, file_offset);
473 entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
474 atomic_inc(&entry->refs);
476 mutex_unlock(&tree->mutex);
481 * After an extent is done, call this to conditionally update the on disk
482 * i_size. i_size is updated to cover any fully written part of the file.
484 int btrfs_ordered_update_i_size(struct inode *inode,
485 struct btrfs_ordered_extent *ordered)
487 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
488 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
492 struct rb_node *node;
493 struct btrfs_ordered_extent *test;
495 mutex_lock(&tree->mutex);
496 disk_i_size = BTRFS_I(inode)->disk_i_size;
499 * if the disk i_size is already at the inode->i_size, or
500 * this ordered extent is inside the disk i_size, we're done
502 if (disk_i_size >= inode->i_size ||
503 ordered->file_offset + ordered->len <= disk_i_size) {
508 * we can't update the disk_isize if there are delalloc bytes
509 * between disk_i_size and this ordered extent
511 if (test_range_bit(io_tree, disk_i_size,
512 ordered->file_offset + ordered->len - 1,
513 EXTENT_DELALLOC, 0)) {
517 * walk backward from this ordered extent to disk_i_size.
518 * if we find an ordered extent then we can't update disk i_size
521 node = &ordered->rb_node;
523 node = rb_prev(node);
526 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
527 if (test->file_offset + test->len <= disk_i_size)
529 if (test->file_offset >= inode->i_size)
531 if (test->file_offset >= disk_i_size)
534 new_i_size = min_t(u64, entry_end(ordered), i_size_read(inode));
537 * at this point, we know we can safely update i_size to at least
538 * the offset from this ordered extent. But, we need to
539 * walk forward and see if ios from higher up in the file have
542 node = rb_next(&ordered->rb_node);
546 * do we have an area where IO might have finished
547 * between our ordered extent and the next one.
549 test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
550 if (test->file_offset > entry_end(ordered)) {
551 i_size_test = test->file_offset - 1;
554 i_size_test = i_size_read(inode);
558 * i_size_test is the end of a region after this ordered
559 * extent where there are no ordered extents. As long as there
560 * are no delalloc bytes in this area, it is safe to update
561 * disk_i_size to the end of the region.
563 if (i_size_test > entry_end(ordered) &&
564 !test_range_bit(io_tree, entry_end(ordered), i_size_test,
565 EXTENT_DELALLOC, 0)) {
566 new_i_size = min_t(u64, i_size_test, i_size_read(inode));
568 BTRFS_I(inode)->disk_i_size = new_i_size;
570 mutex_unlock(&tree->mutex);
575 * search the ordered extents for one corresponding to 'offset' and
576 * try to find a checksum. This is used because we allow pages to
577 * be reclaimed before their checksum is actually put into the btree
579 int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u32 *sum)
581 struct btrfs_ordered_sum *ordered_sum;
582 struct btrfs_sector_sum *sector_sums;
583 struct btrfs_ordered_extent *ordered;
584 struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
585 struct list_head *cur;
586 unsigned long num_sectors;
588 u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
591 ordered = btrfs_lookup_ordered_extent(inode, offset);
595 mutex_lock(&tree->mutex);
596 list_for_each_prev(cur, &ordered->list) {
597 ordered_sum = list_entry(cur, struct btrfs_ordered_sum, list);
598 if (offset >= ordered_sum->file_offset) {
599 num_sectors = ordered_sum->len / sectorsize;
600 sector_sums = ordered_sum->sums;
601 for (i = 0; i < num_sectors; i++) {
602 if (sector_sums[i].offset == offset) {
603 *sum = sector_sums[i].sum;
611 mutex_unlock(&tree->mutex);
612 btrfs_put_ordered_extent(ordered);
618 * taken from mm/filemap.c because it isn't exported
620 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
621 * @mapping: address space structure to write
622 * @start: offset in bytes where the range starts
623 * @end: offset in bytes where the range ends (inclusive)
624 * @sync_mode: enable synchronous operation
626 * Start writeback against all of a mapping's dirty pages that lie
627 * within the byte offsets <start, end> inclusive.
629 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
630 * opposed to a regular memory cleansing writeback. The difference between
631 * these two operations is that if a dirty page/buffer is encountered, it must
632 * be waited upon, and not just skipped over.
634 int btrfs_fdatawrite_range(struct address_space *mapping, loff_t start,
635 loff_t end, int sync_mode)
637 struct writeback_control wbc = {
638 .sync_mode = sync_mode,
639 .nr_to_write = mapping->nrpages * 2,
640 .range_start = start,
644 return btrfs_writepages(mapping, &wbc);
648 * taken from mm/filemap.c because it isn't exported
650 * wait_on_page_writeback_range - wait for writeback to complete
651 * @mapping: target address_space
652 * @start: beginning page index
653 * @end: ending page index
655 * Wait for writeback to complete against pages indexed by start->end
658 int btrfs_wait_on_page_writeback_range(struct address_space *mapping,
659 pgoff_t start, pgoff_t end)
669 pagevec_init(&pvec, 0);
671 while ((index <= end) &&
672 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
673 PAGECACHE_TAG_WRITEBACK,
674 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
677 for (i = 0; i < nr_pages; i++) {
678 struct page *page = pvec.pages[i];
680 /* until radix tree lookup accepts end_index */
681 if (page->index > end)
684 wait_on_page_writeback(page);
688 pagevec_release(&pvec);
692 /* Check for outstanding write errors */
693 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
695 if (test_and_clear_bit(AS_EIO, &mapping->flags))